Facility energy savings or changes in energy use intensity from SEM should be estimated by comparing the facility’smetered energy use(or energy use intensity) during the reporting period with the facility’sadjusted baselineduring the same period—what its energy use (or energy use intensity) would have been had SEM not been implemented. The adjusted baseline is a counterfactual, and it must be estimated usingbaseline perioddata.

Figure 1 illustrates the estimation of SEM energy savings, showing both metered energy consumption and the adjusted baseline. Savings are shown as the dotted area between the adjusted baseline and metered energy consumption. For simplicity, this example does not differentiate between SEM capital projects, operations, maintenance, and behavioral measures.

The adjusted baseline should be estimated using facility energy use data from the baseline period, which should not reflect the SEM program impacts the evaluator wishes to measure. Typically, the baseline period precedes the facility’s SEM implementation.

Using regression, the evaluator should adjust baseline energy consumption for differences between the baseline and reporting periods in output, weather, occupancy, or other measured variables affecting the facility’s energy consumption. Section 4 of this protocol describes two specific regression methods for estimating the adjusted baseline and savings.

This approach for evaluating facility savings from SEM programs will yield accurate savings estimates if the following conditions are met:

• No omitted variable bias (no confounding variables): The regression does not omit any key variables affecting energy use. Specifically, the model controls for all variables that affected energy use and that were correlated with SEM implementation.

• No measurement error: The model independent variables were not measured with error;

For example, omitted variables could bias the SEM savings estimates if: an industrial facility experiences a degradation in the quality of production inputs during SEM, causing energy use per unit of output to increase; and the change in input quality is not accounted for. The change in input quality would be a confounding factor, causing downward bias in the estimated savings.

The evaluator should take steps to minimize the potential for omitted variables and measurement error. These include collecting data on the principal factors affecting facility energy use and conducting statistical tests addressing whether the conditions required for unbiased estimates hold.

SEM may involve implementation of OM&B measures and capital projects, and evaluators may wish to isolate savings from OM&B measures. Section 3 of this protocol discusses estimation of these savings.

For some facilities, it may be necessary for the evaluator to makead hocadjustments to the baseline to capture impacts on energy use that cannot be modeled statistically. These are referred to as “non-routine” adjustments (IPMVP 2012). Section 4 of this protocol discusses the use of non-routine adjustments.

To estimate SEM program energy savings, evaluators should follow these steps:

1. Develop research design (includes sample design)

2. Collect and prepare required data

3. Define baseline and reporting periods

4. Specify regression model

5. Estimate regression model

6. Estimate and document savings

7. Report results

The remainder of this section discusses each of these steps.